DreamFlow: Local Navigation Beyond Observation Via Conditional Flow Matching in the Latent Space
Jiwon Park, Dongkyu Lee, I Made Aswin Nahrendra, Jaeyoung Lim, Hyun Myung
AI summary
Problem
Local navigation in cluttered environments frequently fails when robots become trapped in local minima due to limited onboard sensing and unobserved obstacles beyond their field of view.
Approach
The framework integrates a conditional flow matching module into a deep reinforcement learning policy to probabilistically predict extended environmental representations from limited local height maps, conditioned on navigation context.
Key results
- Introduced DreamFlow, a DRL-based local navigation framework that extends perceptual range via latent conditional flow matching.
- Developed a CFM-based latent prediction module that maps local height map features to extended spatial representations conditioned on navigation context.
- Demonstrated superior latent prediction accuracy and navigation success rates over baselines in simulation.
- Validated real-world deployment on a quadrupedal robot in cluttered environments with frequent local minima.
Why it matters
Provides a robust solution for autonomous mobile systems to navigate highly occluded spaces, advancing capabilities in search, rescue, and exploration robotics.
Abstract
Local navigation in cluttered environments of- ten suffers from dense obstacles and frequent local minima. Conventional local planners rely on heuristics and are prone to failure, while deep reinforcement learning (DRL)-based approaches provide adaptability but are constrained by limited onboard sensing. These limitations lead to navigation failures because the robot cannot perceive structures outside its field of view. In this paper, we propose DreamFlow, a DRL-based local navigation framework that extends the robot’s perceptual hori- zon through conditional flow matching (CFM). The proposed CFM-based prediction module learns probabilistic mapping between local height map latent representation and broader spatial representation conditioned on navigation context. This enables the navigation policy to predict unobserved environ- mental features and proactively avoid potential local minima. Experimental results demonstrate that DreamFlow outperforms existing methods in terms of latent prediction accuracy and navigation performance in simulation. The proposed method was further validated in cluttered real-world environments with a quadrupedal robot. The project page is available at https://dreamflow-icra.github.io.